FlowingFlowing
← Back to all development cases
This case was judged by an anonymized AI reviewer and has not been independently reviewed by people; inspect the draft, cards, and both outputs to assess it.
AutocompleteWith cards better142 / 280 · beu-1029d8e2251745f8

Predictive Coding: a Theoretical and Experimental Review

Quantitative biology · 2107.12979v4

FLOWING EVIDENCE BENCHMARK

How do we compare writing with and without cards?

At the same writing position, the same model produces one first result with cards and one without. All 280 pairs enter the overall anonymized review, which assesses whether a decision not to continue fits the draft.

Overall: 138 with-card wins, 26 ties, 107 no-card wins, 5 pairs where neither was usable, and 4 inconclusive judgments. Both sides wrote text in 218 pairs; 48 had a deliberate stop and 14 included an invalid output.

SAME DRAFT · WITH OR WITHOUT CARDS

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED VERSIONS

SHARED BY BOTH

Draft, writing task, and generation model

A · With cards

Relevant literature cards supplied

B · Without cards

No literature cards supplied

Generation model

Same model and version

A → first output

B → first output

Anonymized AI judge

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Review again with the answer order swapped

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The AI judge sees the same draft and both first outputs without knowing which received cards.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are positions and cards selected?

We selected 40 writing positions in each of seven fields, for 280 comparisons. The position and relevant literature cards were set before each comparison, so both versions began from the same draft.

The main result retains every frozen position, including places where cards might not add a useful next move. This avoids selecting cases based on their outputs, but 49% describes the whole cohort, not a win rate where evidence clearly had an opportunity to help. This cohort did not record that opportunity label for each position before generation, so those two groups cannot yet be separated reliably.

CARD USE · 140/280

Reported card use in continuations with text

The model reports whether a card affected its continuation. In 140 of 280 tests it produced continuation text and reported card use. The previous count of 150 included 10 cases without continuation text and has been corrected.

COMPARISON · 70 / 7 / 30

Which written version was better?

Of the 107 reported-use cases where both versions produced text, the blinded AI judge preferred the with-card version 70 times, found 7 about the same, and preferred the without-card version 30 times.

How are the two table percentages calculated?

ADOPTION COVERAGE

140reported use with text
÷
280all positions
=
50%

WITH-CARD VERSION PREFERRED

70with cards better
÷
107comparable tests
=
65%

The percentages use different groups: 50% counts continuations with text reporting card use among all 280 tests; 65% covers the 107 such cases with text on both sides. The generating model reports card use, while an anonymized AI judge compares writing quality.

Card use is self-reported by the generator. The anonymized AI judge sees the draft and source snippets, so source-aware preference remains possible; there was no independent human review. The overall result describes these 280 frozen positions, not a win probability for any writing task.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

Predictive Coding: a Theoretical and Experimental Review

5 Discussion and Future Directions

…large-scale simulations of predictive coding networks with simultaneous learning has not yet been done. Finally, the implementation of precision in the brain remains largely mysterious. There are competing theories of precision being implemented through lateral connectivity ( K. Friston (2005) ) , feedback superficial-to-superficial connectivity ( Shipp (2016) ) , or primarily through subcortical processes and regions such as the pulvinar ( Kanai et al. (2015) ) . However, to our knowledge, there has been no real explicit experimental tests of the differing predictions made by these theories.

…orithm. Finally, while there has been considerable progress in predictive coding networks setup for the machine learning paradigm of hierarchical static image classification, there has been considerably little work experimenting with potentially more brain-like paradigm such as spatial predictive coding between pixels in an image, or temporal predictive coding over sequences. Indeed, there has not yet even been any large scale study of the performance characteristics of generalized coordinates on spatio-temporal video data, so precisely how effective they are in such settings remains unknown.

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionOverall AI review: With cards better

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Read overall AI review rationale

A = without cards · B = with cards

Both outputs provide coherent academic continuations, but B fits the draft's argument trajectory significantly better. The draft's final paragraph moves from precision implementation to machine learning applications of predictive coding, specifically noting gaps in: (1) brain-like paradigms (spatial/temporal predictive coding), and (2) generalized coordinates on video data. Output B directly extends this thread by addressing the relationship between predictive coding networks and backpropagation—precisely the machine learning paradigm just discussed. The sources confirm this is an active research area with unresolved questions about whether inference learning reaches the same solution as BP after many updates and at comparable cost, matching the draft's pattern of identifying specific technical gaps. Output A pivots to Bayesian sampling vs. variational inference, which is a valid topic but breaks the established trajectory; the draft has not discussed sampling-based inference, and this introduces a new theoretical framework rather than continuing the ML/brain-like paradigms thread. B maintains topical continuity and appropriate scope, while A's scope shift would require more contextual setup to avoid disjunction.

Evidence

Same model and position · retrieved passages supplied · first output

Preferred by blind review

A further open question concerns the relationship between predictive coding networks and backpropagation: although inference learning has been shown to approximate backpropagation under specific conditions, whether it reaches the same solution after many weight updates, and at comparable computational cost, remains unresolved.

Passages supplied to the Evidence version

Card 1 · Deep Predictive Coding Network for Object Recognition

Based on the predictive coding theory in neuroscience, we designed a bi-directional and recurrent neural net, namely deep predictive coding networks (PCN). It has feedforward, feedback, and recurrent connections. Feedback connections from a higher layer carry the prediction of its lower-layer repres…

Read full passage excerpt

Based on the predictive coding theory in neuroscience, we designed a bi-directional and recurrent neural net, namely deep predictive coding networks (PCN). It has feedforward, feedback, and recurrent connections. Feedback connections from a higher layer carry the prediction of its lower-layer representation; feedforward connections carry the prediction errors to its higher-layer. Given image input, PCN runs recursive cycles of bottom-up and top-down computation to update its internal representations and reduce the difference between bottom-up input and top-down prediction at every layer. After multiple cycles of recursive updating, the representation is used for image classification. With benchmark data (CIFAR-10/100, SVHN, and MNIST), PCN was found to always outperform its feedforward-only counterpart: a model without any mechanism for recurrent dynamics. Its performance tended to improve given more cycles of computation over time. In short, PCN reuses a single architecture to recursi …

Card 2 · Predictive Coding Can Do Exact Backpropagation on Convolutional and Recurrent Neural Networks

Predictive coding networks (PCNs) are an influential model for information processing in the brain. They have appealing theoretical interpretations and offer a single mechanism that accounts for diverse perceptual phenomena of the brain. On the other hand, backpropagation (BP) is commonly regarded t…

Read full passage excerpt

Predictive coding networks (PCNs) are an influential model for information processing in the brain. They have appealing theoretical interpretations and offer a single mechanism that accounts for diverse perceptual phenomena of the brain. On the other hand, backpropagation (BP) is commonly regarded to be the most successful learning method in modern machine learning. Thus, it is exciting that recent work formulates inference learning (IL) that trains PCNs to approximate BP. However, there are several remaining critical issues: (i) IL is an approximation to BP with unrealistic/non-trivial requirements, (ii) IL approximates BP in single-step weight updates; whether it leads to the same point as BP after the weight updates are conducted for more steps is unknown, and (iii) IL is computationally significantly more costly than BP. To solve these issues, a variant of IL that is strictly equivalent to BP in fully connected networks has been proposed. In this work, we build on this result by sh …

Card 3 · PredNet and Predictive Coding: A Critical Review

PredNet, a deep predictive coding network developed by Lotter et al., combines a biologically inspired architecture based on the propagation of prediction error with self-supervised representation learning in video. While the architecture has drawn a lot of attention and various extensions of the mo…

Read full passage excerpt

PredNet, a deep predictive coding network developed by Lotter et al., combines a biologically inspired architecture based on the propagation of prediction error with self-supervised representation learning in video. While the architecture has drawn a lot of attention and various extensions of the model exist, there is a lack of a critical analysis. We fill in the gap by evaluating PredNet both as an implementation of the predictive coding theory and as a self-supervised video prediction model using a challenging video action classification dataset. We design an extended model to test if conditioning future frame predictions on the action class of the video improves the model performance. We show that PredNet does not yet completely follow the principles of predictive coding. The proposed top-down conditioning leads to a performance gain on synthetic data, but does not scale up to the more complex real-world action classification dataset. Our analysis is aimed at guiding future research …

Card 4 · Relaxing the Constraints on Predictive Coding Models

Predictive coding is an influential theory of cortical function which posits that the principal computation the brain performs, which underlies both perception and learning, is the minimization of prediction errors. While motivated by high-level notions of variational inference, detailed neurophysio…

Read full passage excerpt

Predictive coding is an influential theory of cortical function which posits that the principal computation the brain performs, which underlies both perception and learning, is the minimization of prediction errors. While motivated by high-level notions of variational inference, detailed neurophysiological models of cortical microcircuits which can implements its computations have been developed. Moreover, under certain conditions, predictive coding has been shown to approximate the backpropagation of error algorithm, and thus provides a relatively biologically plausible credit-assignment mechanism for training deep networks. However, standard implementations of the algorithm still involve potentially neurally implausible features such as identical forward and backward weights, backward nonlinear derivatives, and 1-1 error unit connectivity. In this paper, we show that these features are not integral to the algorithm and can be removed either directly or through learning additional set …

Model-reported card use

Card 2 · fact

Affected output: whether it reaches the same solution after many weight updates, and at comparable computational cost, remains unresolved

Model explanation: This row supplied the specific unresolved question about multi-step convergence and computational cost that the sentence raises as an open problem.

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

A further open question concerns the relationship between predictive coding and other accounts of cortical function, such as the hypothesis that the brain implements approximate Bayesian inference through sampling rather than through the optimisation of a variational free energy bound.

Earlier card-use comparison

The earlier diagnostic compared only cases with reported card use and text on both sides; its scope differs from the overall AI review above.

The blinded model review preferred the Evidence version. Its original rationale appears below.

Read the original model judgment

Output B is grounded in the sources and draft. Source 2 ('Predictive Coding Can Do Exact Backpropagation on Convolutional and Recurrent Neural Networks') explicitly discusses that inference learning (IL) approximates BP with 'whether it leads to the same point as BP after the weight updates are conducted for more steps is unknown' and that 'IL is computationally significantly more costly than BP'—matching B's claims about whether the same solution is reached after many updates and about computational cost. The draft itself discusses 'large-scale simulations of predictive coding networks with simultaneous learning has not yet been done' and gaps in experimental work, making the backpropagation comparison a coherent continuation. Output A introduces 'approximate Bayesian inference through sampling rather than through the optimisation of a variational free energy bound'—a claim about Bayesian sampling that is not mentioned in any source or the draft. While predictive coding's variational free energy framework is broadly related, the specific contrast with 'sampling' accounts is unsupported by the provided materials.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.